7/31/2019 J ANIM SCI-2001-Mills-1584-97
1/16
J A Mills, J Dijkstra, A Bannink, S B Cammell, E Kebreab and J Francedairy cow: model development, evaluation, and application.
A mechanistic model of whole-tract digestion and methanogenesis in the lactating
2001, 79:1584-1597.J ANIM SCI
http://jas.fass.org/content/79/6/1584the World Wide Web at:
The online version of this article, along with updated information and services, is located on
www.asas.org
by guest on June 4, 2012jas.fass.orgDownloaded from
http://jas.fass.org/http://jas.fass.org/http://jas.fass.org/http://jas.fass.org/7/31/2019 J ANIM SCI-2001-Mills-1584-97
2/16
A mechanistic model of whole-tract digestion and methanogenesis in thelactating dairy cow: Model development, evaluation, and application1
J. A. N. Mills*,2, J. Dijkstra, A. Bannink, S. B. Cammell*, E. Kebreab*, and J. France*
*University of Reading, Centre for Dairy Research, Department of Agriculture, Earley Gate, P.O. Box 236,
Reading RG6 6AT, U.K.; WIAS Animal Nutrition, Wageningen University, Marijkeweg 40, P.O. Box 338,
6700 AH Wageningen, The Netherlands; and Institute for Animal Science and Health,P.O. Box 160, NL-8200 AD Lelystad, The Netherlands
ABSTRACT: Dietary intervention to reduce meth-
ane emissions from lactating dairy cattle is both envi-
ronmentally and nutritionally desirable due to the im-
portance of methane as a causative agent in global
warming and as a significant loss of feed energy. Reli-
able prediction systems for methane production over a
range of dietary inputs could be used to develop novel
dietary regimes for the limitation of feed energy loss to
methane. This investigation builds on previous at-tempts at modeling methanogenesis and involves the
development of a dynamic mechanistic model of whole-
rumen function. The model incorporates modifications
to certain ruminal fermentation parameters and the
addition of a postruminal digestive element. Regression
analysis showed good agreement between observed and
predicted results for experimental data taken from the
Key Words: Methane, Animal Models, Dairy Cows
2001 American Society of Animal Science. All rights reserved. J. Anim. Sci. 2001. 79:15841597
Introduction
In 1997, international action was taken in the Kyoto
Protocol to focus on the stabilization of atmospheric
levels of six greenhouse gases, including methane (Mor-
ard, 1999). Methane, as a contributor to global warm-
ing, is second only to carbon dioxide. The majority of
methane production from livestock is associated with
ruminants. Indeed, cattle account for about 73% of the
80 Tg (1 Tg = 1 million metric tons) produced by live-
stock each year (Gibbs and Johnson, 1994). Methano-
genesis also represents a loss of feed energy, and the
1Financial support provided by MAFF Environment [CC0239] is
appreciated by the authors. The research of Jan Dijkstra has been
made possible by a fellowship from the Royal Netherlands Academy
of Arts and Sciences.2Correspondence: phone: 44 (0)118 9316783; E-mail: j.a.n.mills@
reading.ac.uk.
Received August 8, 2000.
Accepted January 8, 2001.
1584
literature (r2 = 0.76, root mean square prediction error= 15.4%). Evaluation of model predictions for experi-mental observations from five calorimetry studies (67observations) with lactating dairy cows at the Centrefor Dairy Research, in Reading, U.K., shows an under-prediction (2.1 MJ/d) of methane production (r2 = 0.46,root mean square prediction error = 12.4%). Applicationof the model to develop diets for minimizing methano-
genesis indicated a need to limit the ratio of lipogenicto glucogenic VFA in the rumen and hindgut. This maybe achieved by replacing soluble sugars in the concen-trate with starch or substituting corn silage for grasssilage. On a herd basis, the model predicted that in-creasing dietary energy intake per cow can minimizethe annual loss of feed energy through methane produc-tion. The mechanistic model is a valuable tool for pre-dicting methane emissions from dairy cows.
potential to increase the ME value of dairy cow dietsthrough a reduction in methane production is sig-nificant.
Calorimetry studies undertaken at the Centre forDairy Research (CEDAR, Reading, U.K.) have pro-duced a database containing simultaneous measure-ment of methane emissions and dietary inputs for dairycows. This investigation focuses on the ability of a modi-fied version of the Dijkstra et al. (1992) mechanisticrumen model to quantify methane production from lac-tating dairy cows. The first objective was to constructa methane module within the existing rumen modelthat allows prediction of methane emissions over arange of dietary inputs. The second objective was toapply the model in the development of novel dietarystrategies for reducing methane emissions.
Materials and Methods
Model Development
Recently, Benchaar et al. (1998b) used both empiricalregression equations and mechanistic approaches to
by guest on June 4, 2012jas.fass.orgDownloaded from
http://jas.fass.org/http://jas.fass.org/http://jas.fass.org/http://jas.fass.org/7/31/2019 J ANIM SCI-2001-Mills-1584-97
3/16
Modeling Methane Emissions from Dairy Cows 1585
simulate methane emissions by dairy cows. Two dy-
namic mechanistic models of ruminant digestion (Bald-
win et al. 1987; Dijkstra et al., 1992) were evaluated
alongside the regression equations for their ability to
predict published values of methane emissions. The
model of whole-rumen function described by Dijkstra
et al. (1992) existed only on the basis of carbon and
nitrogen fluxes. Therefore, it was modified to incorpo-
rate the framework for methane production alreadypresent within the Baldwin et al. (1987) model. Ben-
chaar et al. (1998b) showed an improved prediction of
methane production with both mechanistic models in
comparison with regression equations. However, Ben-
chaar et al. (1998b) presented the Dijkstra et al. (1992)
model as the most reliable model for simulating meth-
ane production over the range of diets tested, with the
model of Baldwin et al. (1987) displaying a larger error
of prediction (36.93% vs 19.87% of the observed mean).
In a comment on the investigation by Benchaar et
al. (1998b), Donovan and Baldwin (1998) suggestedthat
there was no systematic error of prediction for methane
emissions when using the Baldwin et al. (1987) model.Donovan and Baldwin (1998) suggest that the use of
incorrect input parameters within the study by Ben-
chaar et al. (1998b) was a major source of error. Indeed,
Donovan and Baldwin (1998) demonstrated a tendency
for slight underprediction of methane emissions in com-
parison to a significant overprediction shown by Ben-
chaar et al. (1998b). Benchaar et al. (1998a) suggest
that the conflicting results are primarily a result of
uncertain input parameters within the mechanistic ap-
proach to rumen modeling.
Further improvement of the representation of metha-
nogenesis within the Dijkstra et al. (1992) model maybe possible. In particular, and as acknowledged by Ben-
chaar et al. (1998b), the use of the Dijkstra et al. (1992)
rumen model to predict experimental observations of
whole-animal methane production does not account for
postruminal fermentation. In sheep, 8 to 16% of total
methane emissions may result from fermentation
within the cecum and colon (Murray et al., 1976; Mur-
ray et al., 1978). Failure to account for hindgut metha-
nogenesis might account for underestimation by ex-
isting models.
A revised representation of methanogenesis designed
specifically for incorporation into the Dijkstra et al.
(1992) rumen model is described below, followed by amechanistic framework for the simulation of postrumi-
nal fermentation. Unless otherwise stated, the struc-
ture of the rumen model is as presented by Dijkstra et
al. (1992). The general framework for predicting meth-
ane production is a modification of that described by
Baldwin (1995). Excess hydrogen produced during fer-
mentation is partitioned between its use for microbial
growth, biohydrogenation of unsaturated fatty acids,
and the production of glucogenic VFA. The assumption
is made that the remaining hydrogen is used solely and
completely for methanogenesis. This approach treats
hydrogen as a zero pool within the model (France et
al., 1992).
The Production of Hydrogen During Fermentation.Calculation of hydrogen produced during the produc-
tion of the lipogenic VFA, acetate and butyrate, follow-
ing the fermentation of carbohydrate and protein in the
rumen (PHyferm) is as follows:
PHyferm (mol H2/d) = (PAc 2) + (PBu 2)
where PAc and PBu are the amounts (mol/d) of acetate
and butyrate produced during fermentation. Therefore,
2 m o l o f H2 are produced per mole of acetate or butyrate
(Baldwin, 1995).
Further hydrogen is produced as the microbial popu-
lation utilizes amino acids rather than nonprotein ni-
trogen (NPN) for growth (PHyMg). In using the repre-
sentation of methanogenesis described by Baldwin et
al. (1987), Benchaar et al. (1998b) assumed parameter
values for hydrogen flux during microbial growth that
are similar to those presented by Baldwin et al. (1987).
However, the microbial growth calculations within theDijkstra et al. (1992) rumen model should be considered
in terms of polysaccharide-free microbial dry matter.
For this reason, the constants for hydrogen flux during
microbial growth on preformed amino acids or NPN
were reevaluated using the balance equations pre-
sented by Reichl and Baldwin (1975) and Baldwin
(1995). During these calculations, the chemical compo-
sition of the microbial matter was assumed to be as
presented in table 4 of Dijkstra et al. (1992). The revised
requirement for hydrogen during microbial growth
without preformed amino acids was 0.41 mol hydrogen
per kilogram of microbes in comparison to 2.71 mol/kg,
as used by Benchaar et al. (1998b). During growth on
amino acids, there is a net production of 0.58 mol hydro-
gen per kilogram microbes in comparison to 0.42 mol/
kg as used by Benchaar et al. (1998b); therefore
PHyMg (mol H2/d) = Mgaa 0.58
where Mgaa is the quantity of microbial matter (kg
DM/d) produced from growth on amino acids and 0.58
is the yield of H2 per gram of microbial matter.
The Utilization of Hydrogen During Fermentation.The calculation of hydrogen utilization during the pro-
duction of the glucogenic VFA, propionate and valerate,following the fermentation of carbohydrate and protein
in the rumen (UHyferm) is as follows:
UHyferm (mol H2/d) = PPr + PVl
where PPr and PVl are the amounts (mol/d) of propio-
nate and valerate produced during fermentation.
Therefore, there is a net utilization of 1 mol H 2/mol of
propionate and valerate (Baldwin, 1995). The calcula-
tion of hydrogen utilization following growth of mi-
crobes on NPN (UHyMg) is as follows:
by guest on June 4, 2012jas.fass.orgDownloaded from
http://jas.fass.org/http://jas.fass.org/http://jas.fass.org/http://jas.fass.org/7/31/2019 J ANIM SCI-2001-Mills-1584-97
4/16
Mills et al.1586
UHyMg (mol H2/d) = MgNPN 0.41
where MgNPN is the quantity of microbial matter (kg/
d) produced from growth on NPN and 0.41 is the re-
quirement for H2 per gram of microbial matter (see
hydrogen production section). The calculation of hydro-
gen utilization following the biohydrogenation of in-
gested lipid (UHyLi) is done with the equation
UHyLi (mol H2/d) = PLi Liferm 1.805 2.0
where PLi is the intake of feed lipid (mol/d), Liferm is
the proportion of feed lipid subject to lipolysis within
the rumen, 1.805 is the coefficient describing the moles
of unsaturated fatty acid per mole of feed lipid, and 2.0
is the moles of hydrogen utilized per mole of unsatu-
rated fatty acid (Baldwin, 1995).
Hydrogen Balance and Methane Yield in the Rumen.The excess hydrogen available for methanogenesis (Hy)
is calculated as
Hy (mol H2/d) = PHyferm + PHyMg UHyferm UHyMg UHyLi
Ruminal methane production is calculated using the
equation
PCH4 (mol/d) = Hy / 4.0
where 4.0 is the moles of H2 required for the production
of 1 mol of methane resulting from the reduction of
CO2. The calculation of methane energy is made with
PCH4 (MJ/d) = PCH4 (mol/d) 0.883
where 0.833 is the GE of methane in megajoules per
mole.
Additional Dietary Inputs. The Dijkstra et al. (1992)rumen model ignores the contribution of glycerol from
the lipolysis of dietary trigylceride in the fermentation
process. However, in order to maintain a hydrogen bal-
ance within the model, the glycerol is assumed to enter
the amylolytic hexose pool. Pectins are assumed to en-
ter the soluble carbohydrate pool.
New Stoichiometry. Due to the association betweenVFA production and hydrogen metabolism in the ru-
men, accurate simulation of VFA molar proportions isneeded for reliable methane prediction. In the original
model (Dijkstra et al., 1992), the molar proportions of
rumen VFA were based on the stoichiometric data of
Murphy et al. (1982). However, there is a need for re-
finement of these coefficients (Bannink et al., 1997).
Bannink et al. (2000) have recently developed a new
stoichiometry for fermentation within the rumen based
entirely on experimental observations with lactating
dairy cows. Therefore, the rumen model of Dijkstra et
al. (1992) has been modified to accommodate these stoi-
chiometric coefficients.
Modification to Rumen pH Parameters. The originalrumen model (Dijkstra et al., 1992) requires the mean
rumen pH to be specified as an input, based on data
reported in the literature. In conjunction with the mean
pH value, the minimum daily pH and time below a
critical pH (pH 6.3) are required due to the influence
of these parameters on absorptive processes, fiber deg-
radation, and microbial recycling within the rumen.
Bannink et al. (1997) described these pH values as un-certain inputs within the model and showed significant
variation in duodenal NDF flow following a sensitivity
analysis that involved manipulation of these parame-
ters within acceptable limits. For the purposes of this
investigation, the model has been modified to allow a
more dynamic determination of pH based on the rela-
tionship between VFA concentration and pH described
by Tamminga and Van Vuuren (1988) (r2 = 0.71).
The following calculates the mean daily rumen pH
(Tamminga and Van Vuuren, 1988):
pH = 7.73 (0.014 cVFA)
where cVFA is the concentration of VFA in the rumen
(mM). For simplicity in the current modeling exercise,
the minimum daily pH (PM) is subsequently estimated
as mean pH minus 5%, although experimental observa-
tions show a range of 1 to 15% of mean pH depending
on feeding frequency (Sutton et al., 1986):
PM = pH (pH 0.05)
where PM is the minimum pH reached during the day.
Time below a critical pH for reduced fiber digestion (TF
h/24 h), pH 6.3 (Erdman, 1988) is calculated as
TF (h/24 h) = (10.59 pH) + 76.82
TF (h/24 h) = 0 if pH 7.2
TF (h/24 h) = 24 if pH 5.0
This relationship yields a linear increase in TF (h/24
h) from 0 at pH 7.2 and aboveto 24 at pH 5.0 and below.
This assumes a twice-daily feeding pattern (Sutton et
al., 1986).
Modeling Postruminal Digestion and Fermentation.To account for postruminal methanogenesis within the
model, the assumption was made that the fermentation
process is similar to that occurring in the rumen. There-
fore, a mechanistic model of large intestinal fermenta-
tion has been constructed using the existing rumen
model as a basis, and the following modifications were
made. Inputs to the large intestine are rumen outflows
modified for small intestinal digestion. The capacity
for small intestinal digestion of starch is assumed to
depend on the degradability and quantity of starch
flowing from the rumen (Nocek and Tamminga, 1991;
Mills et al., 1999b). Therefore, the following relation-
ship described by Nocek and Tamminga (1991) has been
used to produce a coefficient for small intestinal starch
digestibility (SdSi):
by guest on June 4, 2012jas.fass.orgDownloaded from
http://jas.fass.org/http://jas.fass.org/http://jas.fass.org/http://jas.fass.org/7/31/2019 J ANIM SCI-2001-Mills-1584-97
5/16
Modeling Methane Emissions from Dairy Cows 1587
SdSi (%) = 0.728 RES + 87.9
where RES (% starch intake) is rumen escape starch
and includes microbial storage polysaccharides.
The digestibility of microbial amino acids is set at
0.75. This value is lower than that estimated by AFRC
(1993) for postruminal disappearance to allow for deg-
radation in the large intestine. The true digestibility
of undegraded but potentially rumen-degradable feedprotein is set at 0.7 (Palmquist et al., 1993), but the
indigestible protein fraction remains undegraded. The
NDF is assumed to remain undegraded in the small
intestine. The digestibility of feed and microbial lipid
is set at 0.9 (Palmquist et al., 1993) to allow for calcula-
tion of fecal GE. However, lipid metabolism in the large
intestine is ignored due to its limited significance for
methanogenesis following saturation of fatty acids in
the rumen. Hexose escaping rumen fermentation is
completely absorbed within the small intestine.
Quantitative data regarding the passage of digesta
through the large intestine in lactating dairy cows are
extremely limited in availability. The passage of digestais a combination of plug flow and mixing (France et al.,
1993). However, because the large intestinal model is
based around that of the rumen, the passage kinetics
are assumed to be those of a system involving complete
mixing of digesta. The data from the literature suggest
a mean retention time (MRT) of approximately 9 to
13 h (Huhtanen and Vanhatalo, 1997; Vanhatalo and
Ketoja, 1995; Pellikaan et al., 1999). This equates to a
fractional passage rate range of 7.7 to 11.1%/h with
increases associated with increasing DMI. Therefore,
fractional passage rate in the large intestine is calcu-
lated in the model using the following equation:
KpaLi (%/h) = 1/(0.2 DMI + 13)
where DMI is in kilograms per day. This equation yields
a linear increase in fractional passage rate from 9%/h
at DMI of 10 kg to 11%/h at 20 kg DMI/d. Some selective
retention of microbial matter within the cecum (Van
Soest, 1994) is assumed, and microbial passage rates
are set at 0.85 of the digesta passage rate.
For the sake of the modeling exercise, protozoa are
assumed absent from the large intestine (Demeyer,
1991). Large intestinal volume is set at 10% of calcu-
lated rumen volume (Parra, 1978).Dietary Inputs. Where possible during evaluation of
the model, nutritional inputs are derived from analyti-
cal data reported within the experiments. However,
where information is lacking, estimation of certain in-
puts was made according to data published elsewhere
in the literature. For example, the description of the in
situ kinetics of feed carbohydrate and protein, the data
presented by Nocek and Grant (1987), Tamminga et al.
(1990); Van Vuuren et al. (1990), Nocek and Tamminga
(1991), Bosch et al. (1992), and Huhtanen and Vanha-
talo (1997) have been used.
Model Summary. The differential equations of the 34state variables, representing the nutrient and microbial
pools in the rumen and hindgut, are integrated numeri-
cally for a given set of initial conditions and parameter
values. The model was written in the Advanced Contin-
uous Simulation Language (ACSL, 1995). A fourth-or-
der fixed-step-length Runge-Kutta method with an in-
tegration interval of 0.0001 d was used. The results
presented were obtained by running the model until asteady state was achieved.
Model Application
The model was used to simulate digestion and deter-
mine the efficacy of a range of dietary strategies de-
signed to minimize whole-tract methanogenesis. These
strategies were evaluated through examination of
methane emissions as well as digestibility of the major
nutrients and the overall diet metabolizability (ME di-
vided by GE; ME/GE). Digestible energy was simulated
in the model by deducting the calculated GE of the feces
from the calculated GE of the feed. Simulated methaneand urinary energy values were then subtracted from
the DE to yield a calculated ME value for each diet.
Urine energy values were calculated using an empirical
model of urinary excretion combined with the assump-
tion that N excreted in the urine was related to GE.
This was described by the following locally derived lin-
ear relationship:
Urine energy (MJ/d) = 49.461(Urine N, g/d) + 1.695
r2 = 0.86
Nitrogen excretion in the urine was related to nitrogen
intake using the exponential relationship described byCastillo et al. (1999). The methods of dietary interven-
tion for reducing methanogenesis simulated by the
model were increasing DMI, replacing sugar in the con-
centrate with starch, and increasing dietary energy
density through supplementation of grass silage diets.
Model Evaluation
Evaluation of the model was initially performed on
the same data set used by Benchaar et al. (1998b).
Subsequently, data concerning whole-animal methane
emissions from lactating dairy cows involved in calo-
rimetry studies at CEDAR were used to compare modelpredictions with experimental observations. Data from
five calorimetry studies, representing a range of nutri-
tional strategies involving lactating dairy cows, were
used to test model predictions. Animal and diet details
are summarized in Table 1. Unless otherwise stated,
the trials involved twice-daily feeding and milking.
Cows were trained to use respiration chambers before
the experiments began.
In Trial 1, cows were fed a forage mixture of 3:1 (DM
basis) grass silage:corn silage. This forage was fed at
a 1:1 (DM basis) ratio with either a high- or low-starch
by guest on June 4, 2012jas.fass.orgDownloaded from
http://jas.fass.org/http://jas.fass.org/http://jas.fass.org/http://jas.fass.org/7/31/2019 J ANIM SCI-2001-Mills-1584-97
6/16
Mills et al.1588
Table 1. Trial summary for CEDARa calorimetry studies
Trial No. of cows No. of observations Mean BW, kg DMI, kg/db Milk yield, kg/d
1 6 24 600 18.4 26.9
2 4 16 620 19.8 27.0
3 4 7 606 16.3 29.7
4 4 8 671 17.7 21.7
5 3 12 651 20.9 31.1
aCentre for Dairy Research.bDMI corrected for volatile losses.
concentrate as part of a total mixed ration (TMR) and
at ad libitum or a restricted level of intake (85% of ad
libitum DM). Trial 2 involved the feeding of a 1:1 (DM
basis) mixture of high- and low-DM whole-crop wheat
(WCW) silage as described by Sutton et al. (2001). The
WCW mix was fed with first-cut grass silage in a 1:2
(DM basis) grass silage:WCW ratio. The forage was
offered for ad libitum consumption with 8.2 kg DM of
a concentrate mixture. Treatments involved the re-
placement of WCW silage with NaOH treated WCWsilage or by altering the formulation of the concentrates.
Trials 3 and 4 involved feeding fresh grass three times
daily ad libitum. Concentrates were fed at 5.2kg DM/
d. Trial 4 was undertaken 12 mo after Trial 3. Trial 5
involved feeding a 3:1 (DM basis) corn silage:grass si-
lage mixture with experimental procedures presented
by Cammell et al. (2000). The corn silage was harvested
at two stages of maturity, defined by DM content. Four
different concentrates were fed alongside both forage
mixtures. Concentrates varied in starch source and de-
gradability and were fed at 8.7 kg DM/d with the forage
mixture at ad libitum intake.
Statistics. Regression analysis between observed andpredicted values has been used to demonstrate the re-
liability of model predictions. Error of prediction is esti-
mated from the calculation of root mean square predic-
tion error (MSPE) and expressed as a percentage of
the observed mean:
MSPE = (Oi Pi)2/n
where i = 1, 2, . . . n; n is the number of experimental
observations; and Oi and Pi are the observed and pre-
dicted values (Bibby and Toutenburg, 1977). The MSPE
is decomposed into overall bias of prediction, deviation
of regression slope from 1, and the disturbance propor-
tion. These components of MSPE were defined by Ben-
chaar et al. (1998b).
Results
Model Prediction of Literature Data
Results of the regression between observed and pre-
dicted methane production from the literature experi-
ments are shown in Figure 1. In this instance, the model
overestimated average methane production by 0.52 MJ/
d. The r2 was 0.76 and the root MSPE was 15.4% of
the observed mean. The contribution of the disturbance
proportion was 73% of the MSPE. For all simulations,
the mean (and SD) simulated contribution of large in-
testinal fermentation to total methane emissions was
9.14% 2.64.
Model Prediction of CEDAR Trials
The results from one cow (Cow 28, Trial 5) have been
removed from the analysis due to unusually low nutri-ent digestibilities and methane emissions comparedwith contemparies on the same dietary regime. Figure
2 displays the regression analyses between methaneproduction for all five trials (67 observations). Table 2displays the summary statistics for each of the 5 trials,
together with a summary for all observations combined.Overall mean predicted methane production was un-
derestimated by 2.09 MJ/d with an r2 of 0.46 and a rootMSPE of 12.36% of the observed mean. The slope andintercept were different from 1 (P < 0.05) and 0 (P
Recommended